Extraction and Analysis of Knowledge for Automatic Software Repair
Year: 2014
Type: dissertation
Abstract: Bug fixing is a frequent activity in the software life cycle. The activity aims at removing the gap between the expected behavior of a program and what it actually does. This gap encompasses different anomalies such as the failure of a program facing to a given scenario. Bug fixing is a task historically done by software developers. However, in the recent years, several automatic software repair approaches have emerged to automatically synthesize bug fixes. Unfortunately, bug fixing could be even hard and expensive for automatic program repair approaches.In this thesis, we aim at adding repair approaches strategies to optimize the search of solutions in the repair search space. These strategies consume information extracted from repairs done by developers.Then, for validating the repair approaches and our strategies, we focus on the evaluation of automatic repair approaches. We aim at introducing methodologies for defining how researchers can evaluate repair approaches in a meaningful manner. For example, the performance of a repair approach depends on the defect dataset used to evaluate the approach. First, we define a methodology to define defect datasets that minimize the possibility of biased results.We present a dataset that includes a particular kind of defect: if conditional defects. Then, we aim at measuring the repairability of this kind of defect by evaluating three state-of-the-art automatic software repair approaches.
Source: HAL
Author: Martinez, Matias
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7065940335
